For most of the generative AI boom, the two leading frontier labs looked roughly alike from the outside: huge valuations, huger losses, and a shared bet that scale would eventually pay for itself. This week, as both filed or prepared to file for public listings, that resemblance broke. Anthropic told investors it expects its first ever quarterly operating profit, around $559 million on $10.9 billion of revenue in the current quarter. OpenAI, working with Goldman Sachs and Morgan Stanley toward a listing that could value it above a trillion dollars, is asking public markets to fund another four or five years of losses.
The numbers behind the divergence, reported first by the Wall Street Journal and dissected this week in Forbes and The Daily Upside, are striking. Anthropic now spends roughly 56 cents on compute for every dollar of revenue, down from 71 cents a quarter ago. Around 85 percent of its revenue comes from enterprise and developer customers, the kind that pay reliably and run predictable workloads. OpenAI’s mix is the mirror image: roughly 85 percent of its revenue is tied to ChatGPT consumer subscriptions, and about 95 percent of ChatGPT’s users pay nothing. The free tier, in other words, is doing most of the burning.
It is tempting to call this the moment Anthropic won. That would be premature. Gary Marcus, picking through the same numbers, points out that the profit projection appears to rest partly on a one off discount from SpaceX, which is selling Anthropic compute under a $15 billion a year deal that ramps up over time. The S−1 SpaceX filed this week shows reduced payments for May and June. Whether subsequent quarters look anything like Q2 is the open question, and Anthropic itself has told investors it may not be profitable for the full year as compute spending climbs.
What is harder to argue with is the structural difference. Enterprise customers generate three to five times more revenue per token than consumer users, contracts are sticky, and query patterns are predictable enough to plan capacity around. More than 500 companies now pay Anthropic over a million dollars a year, and eight of the Fortune 10 are customers. That is the kind of revenue base that has historically supported public software multiples. OpenAI’s case rests on a different bet: that agentic AI will eventually convert the 900 million weekly ChatGPT users into something that looks more like Salesforce and less like a very expensive free chatbot.
The two companies are also racing on a second front, one that does not show up in the quarterly numbers. Anthropic’s new enterprise joint venture, backed by Blackstone, Hellman & Friedman and Goldman Sachs with reported $1.5 billion in capital, just acquired Fractional AI, an integration firm that had been working in OpenAI’s partner ecosystem. OpenAI is responding by raising over $4 billion for its own services arm, The Deployment Company, valued at around $10 billion. Both labs have concluded that selling API access is necessary but not enough; the margin sits in the consulting layer that installs models inside companies.
That insight, that AI value will accrue partly to whoever owns the deployment pipe, is reshaping more than just the AI labs. The same week, Microsoft and EY announced a $1 billion, five year initiative to embed Microsoft engineers and EY consultants inside enterprise clients, a structure that looks remarkably like what Anthropic and OpenAI are building from the other direction. The big consulting firms appear to have read the same memo, and the AI labs would rather not cede that ground.
For institutional investors about to be handed two S−1s within months of each other, the wager is now legible in a way it was not a year ago. One company is asking the public markets to value it on the promise of an Amazon style flywheel that does not yet exist. The other is showing up with a quarter that, if it holds, looks more like enterprise software than infrastructure. The Amazon comparison may still be right for one of them. It is increasingly clear it cannot be right for both.